← back to timeline

EXP-TRIAD-DOSE33

success

Triad Dose-Response — Does Gradient Compression Operate at Minimum N?

2026-03-10 L2 level paper3 2 runs $0.01
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

We know three agents is the minimum for a pecking order. We know a synthesizer flattens the pecking order among five. Does it work at three — the smallest possible ecosystem? Answer: YES, and it's even STRONGER. One synthesizer among three agents cuts hierarchy by 68% (vs 58% among five). Fewer paths to compress = each compression counts more.

Predictions we made before running

4/4 confirmed

Technical details

Research hypothesis

OPEN QUESTION: "emergence threshold: at what N agents + interaction density do emergent properties appear?"

Known: N=3 is the minimum for trophic hierarchy (EXP-DYAD flat, EXP-ASYM hierarchy TR=0.444). At N=5, 1 synthesizer (20%) suppresses hierarchy massively (TR=0.223 ± 0.057, Cohen's d=5.38 vs baseline).

UNKNOWN: does the gradient compression mechanism operate at N=3?

At N=3, 1 synthesizer = 33% dose (vs 20% at N=5). The triad topology is minimal — only 3 pairwise relationships. The synthesizer cross-pollinates vocabulary between the other two agents, but with fewer agents to connect, the compression effect may be weaker OR stronger (fewer paths to compress).

This experiment starts a SECOND dose-response curve at N=3, enabling cross-scale comparison with the complete N=5 curve.

Replacing Gamma (concrete — most upstream in EXP-ASYM, TR=0.724) with a synthesizer gives the gradient compression the HARDEST test: can it suppress a hierarchy where the replaced persona was the strongest differentiator?

Experimental setup

Type: simple

Condition Parameters
TRIAD_1SYNTH persona: MIXED, interaction: LIVE, n_agents: 3, note: 2 non-integrative + 1 synthesizer. Direct comparison: EXP-ASYM (3 non-integrative)

Factors: n_agents (3) × synthesizer_fraction (0.33)

Parameters

n_agents
3
n_rounds
36
n_runs_per_condition
1
model
gemini-2.5-flash-lite
temperature
0.9
scheduler
round_robin

Trophic Ratios by Condition

Mean trophic ratio per agent across runs. Error bars = ±1 std dev. Higher TR = more upstream (exporter).